The characteristics of probability distribution of groundwater model output based on sensitivity analysis

The probability distribution of groundwater model output is the direct product of modeling uncertainty. In this work, we aim to analyze the probability distribution of groundwater model outputs (groundwater level series and budget terms) based on sensitivity analysis. In addition, two sources of uncertainties are considered in this study: (1) the probability distribution of model's input parameters; (2) the spatial position of observation point. Based on a synthetical groundwater model, the probability distributions of model outputs are identified by frequency analysis. The sensitivity of output's distribution is analyzed by stepwise regression analysis, mutual entropy analysis, and classification tree analysis methods. Moreover, the key uncertainty variables influencing the mean, variance, and the category of probability distributions of groundwater outputs are identified and compared. Results show that mutual entropy analysis is more general for identifying multiple influencing factors which have a similar correlation structure with output variable than a stepwise regression method. Classification tree analysis is an effective method for analyzing the key driving factors in a classification output system.